The bear wants a small green apple and a big one : Insights into anaphoric reconstruction in child and adult Romanian

. We investigate how 4-year-old children reconstruct the anaphoric pronoun unul ‘one’ when modified by Color or Size adjectives in Romanian, an understudied Noun-Adjective language. In a Truth Value Judgment Task, in Ursuleţul vrea un măr verde mic şi unul galben ‘The bear wants a small green apple and a yellow one ’, both adults and children accept unul ‘one’ to refer to apple or small apple . However, in Ursuleţul vrea un măr verde mic şi unul mare ‘The bear wants a small green apple and a big one ’, children take unul ‘one’ to refer to apple or green apple, while adults only accept green apple. Based on these findings, we argue that while adults seem to be sensitive to hierarchical nesting in anaphoric reconstruction, children seem to operate on the basis of other reconstruction preferences at this developmental stage.


Introduction.
Language makes use of multiple ambiguous situations, where the anaphor one can refer to N or an N' (Pearl & Lidz 2006).In the current paper, we investigate how children reconstruct the anaphoric pronoun unul 'one' in Romanian when the referential noun is modified by Color and Size adjectives, and the anaphor is modified by a Color or a Size adjective.
(1) a. Ursuleţul vrea un măr verde mic şi unul galben.bear.the wants an apple green small and one.the.N.SG yellow 'The bear wants a small green apple and a yellow one.' b.Ursuleţul vrea un măr verde mic şi unul mare.bear.the wants an apple green small and one.the.N.SG big 'The bear wants a small green apple and a big one.'While multiple studies have been conducted on anaphoric reconstruction in other languages (such as English: Lidz, Waxman & Freeman 2003;German: Hubert 2009;Italian: Tasinato & Sanfelici 2022), our study is the first to explore anaphoric reconstruction of unul 'one' experimentally in child and adult Romanian.Our goal is to establish whether Romanian children and adults prefer anaphoric reconstruction to N or N', and whether their preferences differ in any way.In (1a), for instance, where the pronoun unul 'one' is modified by the color adjective galben 'yellow', does unul 'one' refer to un măr 'an apple' or un măr mic 'a small apple '? In (1b), where the pronoun unul 'one' is modified by the size adjective mare 'big', does unul 'one' refer to un măr 'an apple' or un măr verde 'a green apple'?The answers to such questions have an important bearing on how we represent adjective modification syntactically (flat/hierarchically).Previewing the results, we show that adults prefer maximal anaphoric reconstruction constrained by hierarchical nesting.That is, in (1b), they prefer the big apple to also be green, but in (1a), the yellow apple can be big or small.Children, however, are more variable in their behavior, sometimes reconstructing the anaphor to N, and other times to N'.

Background.
2.1.THE ACQUISITION OF ANAPHORIC ONE.As shown by Pullum & Scholz (2002), language has both deictic uses of one (2a) and anaphoric uses of one (2b).While the focus of our current study is on anaphoric one, deictic uses of one are quite frequent both in child-directed speech, as well as in child language.
(2) a.The green one is mine.(Pullum & Scholz 2002: 32) b.I'd rather teach linguistics to a student of mathematics than to one of any discipline in humanities.(Pullum & Scholz 2002: 32) Moreover, various studies (e.g.Karmiloff-Smith 1980;Tasinato & Sanfelici 2009) suggest that young children may sometimes interpret anaphoric one deictically instead of linking it to the previously mentioned noun.As shown in (3), children seem to produce anaphoric one from very early on: (3) CHI: there was a frog (.) on the kitchen center (.) in Sams_Club CAR: whoa, there was?CHI: yeah, he was a small one (Weist, Emily, 3;00.12) The literature on the acquisition of anaphoric one has been centered on the question of how children reconstruct its reference.Some studies have claimed that (i) anaphoric one is initially interpreted deictically, others have claimed that (ii) it can refer to either N or N', but that there may be a preference for N'.
Related to the first claim that children may interpret anaphoric one deictically, Avrutin & Coopmans (2000) have argued that children determine possible antecedents relying on the visual context, which makes grammatical connections unnecessary.Karmiloff-Smith (1980) found that, in a story context, children under the age of 6 prefer to interpret definite noun phrases and pronouns deictically rather than anaphorically.Moreover, Tasinato & Sanfelici (2022) have recently argued that Italian children are more accurate with the anaphor one in coordination NP ellipsis (4) than in discourse NP ellipsis (5), where they interpret anaphoric one deictically to a larger extent (see also Figure 1).Interestingly, this interpretation characterizes 3-year-olds but not older children.
(4) The lion washed the red car and the blue one.
(NP coordination ellipsis) (5) The elephant washes the red car.The lion washes the blue one.
(discourse ellipsis) Figure 1.Examples of picture where the 2 nd NP refers to a bicycle not a car (NP coordination ellipsis, discourse ellipsis) Tasinato & Sanfelici (2022) explain this difference by arguing that ellipsis is a type of syntactic dependency whose resolution is strictly conditioned by locality, and that another DP (the lion) acts as an intervenor between the elided NP and its antecedent, thus blocking the anaphoric relation between one and car.Moreover, the acquisition picture seems to suggest that NP ellipsis in NP coordination is acquired before NP ellipsis in discourse.
In contrast to these studies, Wijnen, Roeper & van der Meulen (2003) showed that young children interpret nominal ellipses in an anaphoric manner, based on syntactic structure building principles.They tested young children's understanding of sentences containing nominal ellipsis marked by a bare cardinal, such as two in example (6), by evaluating them against three upside down pictures: one of two children inside the sandbox (target picture), one of two children outside the sandbox, and one of two adults outside the sandbox.4-year-olds answered the target question in (6) with a yes more often for the target picture than for the other pictures.
(6) There are some kids playing in the sand box.Are two upside down?(Wijnen et al. 2003: 3) Hubert ( 2009) also found that children reconstruct anaphoric one in an adult-like manner in both sentence coordination ellipsis contexts (Ele is washing a car and Dino is washing one) and in discourse ellipsis contexts (Ele is washing a car.Is Dino washing one?) from a very young age (3;8).Only 2 of 38 children interpreted one deictically as referring to a bicycle rather than a car.Putting these findings together, it seems that children sometimes reconstruct anaphoric one as deictic, but that this tendency mostly characterizes 3-year-old children.
Related to the second claim that anaphoric one can refer to either N or N', but that there may be a preference for N', it has been argued by Bryant (2006) that anaphoric one reconstructs to phrasal N' rather than N.This point of view has, however, been challenged by Pullum & Scholz (2002), who argued that one may reconstruct to either N or N' (see 2b).Lidz et al. (2003) investigated whether children reconstruct one to N'.To evaluate the role of the input, they conducted a corpus analysis of parental speech (directed at two children) using the CHILDES database (MacWhinney 2000).Their study reveals that in most of the cases (95%), anaphoric one referred to a previously mentioned N or it was unclear whether it referred to N or a larger projection than N. Lidz et al. (2003) take these findings related to the child-directed input to suggest that children establish the reference of anaphoric one on the basis of innate syntactic principles.To see what their innate preference is, they conducted a looking time experiment with 18-month-old infants.After seeing a yellow bottle, infants heard Now look.Do you see another one?They then were presented with two bottles, an identical one of the same color, and one of a different color.Interestingly, infants showed longer looking times for the same color bottle.Pearl & Lidz (2006) argue that learners must already be equipped with the knowledge that one refers to N'.While the findings of the study by Lidz et al. (2003) are quite striking, a possible counterargument to their claims is that what they have shown is that children have knowledge of another one rather than of anaphoric one (Hubert 2009).Hubert (2009) further investigated the anaphoric reconstruction of one in older German native children (4-6 -year-old) in a modified version of the truth value judgement task (Crain & Thornton 1998), looking both at sentence coordination ellipsis and discourse ellipsis to test whether Bryant's (2006) hypothesis that sentence coordination ellipsis disfavors deictic interpretations of the anaphor and favors N'-reconstruction.Participants were presented with pictures and had to say whether they agree with certain sentences or not.Apart from cases where both the antecedent and the anaphor are unmodified, Hubert (2009) investigated multiple conditions (7, 8): where the antecedent is modified by a Color adjective and a Size adjective and the anaphor is unmodified (see 7 a, b), and where the antecedent is modified by a Color adjective and the anaphor is modified by a Size adjective (see 8).
(7) a. Ele is washing a red car/ a big red bike and Dino is washing one.
b. Ele is washing a red car/a big red bike.Is Dino washing one?
(8) Ele is washing a red car and Dino is washing a small one.
Figure 2. Example pictures for (7) Figure 3. Example pictures for ( 8) Hubert (2009) found that children prefer to reconstruct an antecedent that shares the features of the object of the previously mentioned sentence, and this tendency is more pronounced with coordination rather than discourse ellipsis.Interestingly though, when both the antecedent and the anaphor are modified by an adjective (see 8), children reconstruct to N.
Apart from these conditions, Hubert ( 2009) also conducted a pretest on monolingual adult German native speakers testing whether they do N'-reconstruction in contexts where the referential nominal antecedent was modified by two adjectives (a Color adjective and a Size adjective) and the anaphor is modified by an adjective (a Color adjective or a Size adjective).Participants were presented with pictures that (mis)matched the antecedent with respect to one or two properties.They were asked to respond with either 'yes' or 'no' to the question in ( 9). ( 9) Ele is washing a big red car.Is Dino washing a small one?All subjects accepted the picture corresponding to one reconstructing to the N' (red car), and most of them (20/23) also accepted the picture corresponding to one reconstructing to N, mismatching the antecedent in color.In a posttest, Hubert (2009) also tested 8 4-year-olds on sentences such as those in (9) and found that all 8 children accepted a picture showing Dino washing a car with a different color.This suggests that, just as adults, children are able to reconstruct to N. While the small number of child participants might cast doubt on the generalizability of the findings, the fact that 8 children were able to reconstruct to N is telling with respect to the availability of minimal reconstruction in child language.
Overall, the findings about anaphoric one seem to suggest that children have a general preference for maximal reconstruction to N', particularly with coordination ellipsis.However, they are also able to reconstruct to N from a very young age.Of the contexts above, our current study focuses on contexts where the antecedent is modified by Color and Size adjectives and the anaphora is modified by a Color or Size adjective.To gain more insight into the possible challenges related to anaphoric reconstruction, we delve into the domain of adjective orders in Romanian.

ADJECTIVE ORDERS IN ROMANIAN.
A significant portion of adjective research has focused on adjective ordering restrictions (AORs), explored in various theoretical and experimental studies (Dixon 1982;Matthei 1982;Cinque 1994Cinque , 2005Cinque , 2010;;Scott 2002;Bryant 2006;Scontras et al. 2017Scontras et al. , 2019, among others), among others).These discussions often revolve around whether speakers arrange adjectives based on cognitive dimensions and subjectivity or innate syntax.The prevailing view for English suggests a specific hierarchy (see 10), correlated with adjective subjectivity.
(10) QUALITY > SIZE > SHAPE > COLOR > PROVENANCE a beautiful big round red British lamp However, there seems to be crosslinguistic variation in how adjectives are ordered.Adjective orders have been argued to be more flexible in languages such as Greek (Leivada & Westergaard 2019), Hebrew (Trainin & Shetreet 2021) and even German (Grohe & Schulz 2021).
Our study focuses on Romanian, a language that shows a post-nominal adjective order (11c, d), unlike English, where adjectives are prenominal (11a).( 11 There is a longstanding debate in the literature regarding AORs in Romanian: some studies claim that adjective orders are fixed in Romanian (Cinque 1994(Cinque , 2005(Cinque , 2010)), while others argue that adjective orders are flexible (Cornilescu & Giurgea 2013, Cornilescu & Nicolae 2016, Cornilescu & Cosma 2019).According to Cinque (1994Cinque ( , 2005Cinque ( , 2010)), Romance languages, including Romanian, mirror the Germanic adjective in postnominal order.This order is explained through Roll-Up-of-N, a set of movement operations starting from the basic English order.To give an example (12), the NP moves out of its position to an outer specifier (Step 1) of FPcolor, then the newly formed FP containing FPcolor moves out of its position to the outer specifier of the projection hosting FPsize (Step 2).
Recently, however, Bleotu & Luciu (to appear) and Truşcă & Bleotu (to appear) have experimentally investigated whether Romanian adult speakers have different adjectival ordering preferences compared to British English speakers.They found that British English speakers tend to observe the hierarchy in (10), and that they prefer (11a) over (11b), placing the Color adjective closer to the noun than the Size adjective.In contrast, Romanian speakers accepted both orders alike (11c, d).This suggests that a more accurate way of capturing adjective orders in Romanian may be adjunction (Abels & Neeleman 2010, Kremers 2003), which successively merges adjectives to the left as modifiers of the noun (see 12).
Another area of research on adjectives has focused on ordering and interpreting adjectives in terms of sets and subsets.According to recent studies on both English and Romanian adults and children (Bleotu & Roeper 2021a, b;Bleotu & Roeper 2022a, b;Foucault et al. 2022Foucault et al. , 2024;;Bleotu, Foucault & Roeper 2023a, b), in a context where one wants to refer to a subset of objects from a set of objects, a constraint is at work at the syntax-semantics interface (Recursive Set-Subset Ordering Principle (RSSO)), such that Set modifiers are merged earlier to the noun than Subset modifiers: [long/green subset [shortset leaves]].This principle derives from the properties of Merge in Universal Grammar and provides an automatic Set-Subset ([subset [set]]) interpretation to the syntactic hierarchy.This can be captured in the cartographic approach, à la Cinque (1994Cinque ( , 2005Cinque ( , 2010) ) or through an adjunction account (Abels & Neeleman 2010;Kremers 2003) that successively merges Set and Subset adjectives to the right as modifiers of the noun (see ( 12)), where the firstly adjoined adjective may be interpreted as the Set adjective and the adjective further away may be interpreted as the Subset adjective.
(12) Aside from these hierarchical approaches that are based either on cognitive orders or setsubset hierarchies, there exists a third possibility to represent nouns modified by a color and a size adjective: a flat coordinative structure (13), where the two adjectives are not hierarchically ordered with respect to each other: (13) Previous language acquisition studies regarding AORs based on cognitive properties suggest that children may not have innate fixed hierarchies (see Grohe & Schulz 2021) and, importantly, that recursive set-subset hierarchies override them (Bleotu & Roeper 2022a, b).Looking at anaphoric reconstruction relative to nouns modified by Color and Size adjectives allows us to probe into the question of whether adults and children take the adjectives to be placed in a fixed hierarchy and whether they choose to reconstruct N or N'.

Experiment.
We investigate how adults and 4-year-old children who are native speakers of Romanian reconstruct the anaphoric pronoun unul 'one' in a context where the antecedent is modified by a Color and Size adjective and the anaphor is modified by a Color or a Size adjective.
3.1.PREDICTIONS.Overall, given that we tested children older than 4, we do not expect to see many deictic errors, where the anaphor is interpreted as referring to a different referent.Rather, we expect children to be anaphoric in their reconstruction.
Nevertheless, children may exhibit different anaphoric reconstruction patterns than adults.As discussed in the previous section on adjective orders in Romanian, adjectives can be analyzed both under hierarchical approaches (cartographic or adjunction), where nouns modified by Color adjectives recursively embed Size adjectives, as well as under coordination approaches, where Color adjectives are coordinated with Size adjectives, and together they modify nouns.If we assume that adults are sensitive to nesting and anaphoric reconstruction is maximal, then one should reconstruct to the N' under the Size/Color adjective (see Table 1).In Ursuleţul vrea un măr verde mic şi unul mare 'The bear wants a small green apple and a big one', adults should accept unul 'one' to only refer to măr verde 'green apple', consisting of the adjective under the Size adjective and N.More specifically, adults substitute the size adjective mic 'small' modifying the antecedent with the size adjective mare 'big' and then maximally reconstruct unul 'one' to the noun măr 'apple', modified by color adjective verde 'green.'However, in Ursuleţul vrea un măr verde mic şi unul galben 'The bear wants a small green apple and a yellow one', unul 'one' should reconstruct to măr 'apple' under a hierarchical approach -both a big and a small apple should be allowed.When no structural preference constrains the interpretation, it might be the case, nonetheless, that other (possibly cognitive) biases are allowed to occur, e.g., some adults might still prefer to keep the previously mentioned property.For instance, they might prefer the second apple to also be small for parallelism reasons.
It is unclear how children will behave: they might opt (i) for maximal reconstruction (as in Lidz et al. 2003) or (ii) for minimal N reconstruction, accepting that the apple be a different color than green in Ursuleţul vrea un măr verde mic şi unul mare 'The bear wants a small green apple and a big one' (as in Hubert 2009).Children might also associate the two adjectives modifying the noun with a (flat) coordination analysis (Table 2), given that coordination has a privileged status in child language (Bleotu & Roeper 2021a, b;Foucault et al. 2022).This leads to reconstruction to N, given that anaphoric reconstruction only applies to constituents.In contrast, associating the Size and Color adjectives modifying the noun with a hierarchical, nested structure, where color adjectives are merged to the nouns first and size adjectives are merged afterwards, may lead to reconstruction to N'.
Regarding a possible difference between DP coordination ellipsis and sentence coordination ellipsis (Ursuleţul vrea un măr verde mic şi unul mare 'The bear wants a small green apple and a big one' vs. Ursuleţul vrea un măr verde mic şi lupul vrea unul mare 'The bear wants a small green apple and the wolf wants a big one'), it might be that participants prefer maximal reconstruction in both DP coordination ellipsis contexts and in sentence coordination ellipsis contexts, given that both contexts involve coordination (see Hubert 2009).However, it might also be that they prefer maximal reconstruction in DP coordination ellipsis contexts than in sentence coordination ellipsis contexts due to the closeness of the antecedent and the absence of an intervenor in the DP coordination ellipsis contexts (Tasinato & Sanfelici 2022).The control sentences tested participants' reconstruction of the anaphoric pronoun unul 'one.the.M/N.SG'/ una 'one.the.F.SG' when the antecedent was an unmodified nominal (see 15a), and of the pronoun altceva 'something else' (see 15b).The pronouns were each tested both in a situation where the boy gives the animal the right object, and in a situation where the boy gives the animal the wrong object.Mouse wants a dress and Koala wants something else 'Mouse wants a dress and Koala wants something else.' The test sentences had the form Animalul X vrea un N1 Color1 Size1 şi (animalul Y vrea) unul Color2/Size2 'Animal X wants a Size1 Color1 N1 and (animal Y wants) a Color2/Size2 one' (see ( 16) and ( 17)).
( 16) DP-coordination Ursuleţul vrea un măr verde mic şi unul mare /galben.bear wants an apple green small and one big/ yellow 'The bear wants a small green apple and a big/yellow one.' Porcuşorul vrea un măr portocaliu mare şi Oiţa vrea unul mic/roşu.Piggy wants an apple orange big and Sheepy wants one small/ red 'Piggy wants a big orange apple and Sheepy wants a small/red one.' The objects given either matched the referential nominal in Color/Size, matched it partially in Color/Size, or mismatched the nominal referent altogether.Figure 5 exemplifies the procedure.

Figure 5. Example of items involving anaphoric unul 'one' modified by a Size adjective
The bear wants a small green apple and a big one.What does the bear want?
The bear wants a small green apple and a big one.Participants heard a sentence and were asked to repeat it.They were then introduced to an array of objects that had either the size/color of the antecedent or of the anaphor modifier.In this example, they were presented with two apples of the same color (as the antecedent): a small green one, a big green one, two apples of a different color from the antecedent (a small yellow one, a big yellow one) and four bananas (mis)matching the antecedent in size and/or color (a small green banana, a big green banana, a small yellow banana, a big yellow banana).Next, participants witnessed a scene where the boy, Mihai, gave the bear a small green apple, and then they were asked what else they thought the boy should give the bear.Participants then witnessed a scene where the boy gave the bear an object (in this example, the wrong object, i.e., a big green banana).Children were then asked to say whether the boy did the right thing, and to correct him if they thought he had not.
3.4.RESULTS.Both children and adults were quite accurate in the controls.Children were 100% accurate in handling unul/una 'one' when the antecedent was an unmodified noun, while adults were 90.62% accurate.Children were 66.7% accurate in handling altceva 'something else', while adults were 84.75% accurate.Interestingly, the lower accuracy in handling altceva 'something else' was due to participants preferring to give the character a different object than the one that had been given to them.Nonetheless, participants can be considered accurate even with altceva 'something else' given that the object they suggested was always a different one than the antecedent.The two controls establish a baseline for analyzing participant responses: children are anaphoric with unul/una 'one' and avoid co-reference with an antecedent for altceva 'something else'.
Regarding the interpretation of the test sentences, we find that children are anaphoric in their reconstruction rather than deictic, as there were only 2 deictic errors where children accepted other referents in mismatch situations (i.e., where animals received the wrong objects).
However, looking at children's and adults' anaphoric reconstruction in match and partial match situations (where animals received objects that fully/partially matched the sentences they heard) reveals that children and adults differ in their anaphoric reconstruction preferences.(see Figure 6).In Ursuleţul vrea un măr verde mic şi unul mare 'The bear wants a small green apple and a big one', adults maximally accept one as green apple, rejecting reconstruction to apple of a different color.In Ursuleţul vrea un măr verde mic şi unul galben 'The bear wants a small green apple and a yellow one', adults accepted one to refer to either big apple or small apple.In contrast, children split into four groups.Group 1 (N= 2 children) behaved like adults.Group 2 (N =4) rejected reconstruction of unul 'one' to an apple of a different color in The bear wants a small green apple and a big one, as well as to an apple of a different size in The bear wants a small green apple and a yellow one.Group 3 (N=7) accepted that unul 'one' can refer to an apple of a different size/color in both structures.Group 4 (N= 5) displayed mixed non-adult-like behavior, taking unul 'one' to refer to an apple of a different color/size exactly 50% per anaphor type (color or size).
Using R Studio ( 2021), we fitted the data into a generalized linear model (glm) with Answer as a dependent variable, the fixed effects Anaphora (Color vs Size one), Group (Children vs. Adults), and Condition (Match/Mismatch/Partly Match).Firstly, we observed a marginal significance regarding the effect of group on our response variable (F(1, 34) = 3.222, p = 0.0732).Additionally, the effect of condition emerged as highly significant (F(2, 34) = 623.947,p < 2e-16), indicating a robust impact of the conditions on our response variable.Moreover, the presence of distinct anaphora types significantly affected our response variable (F(1, 34) = 9.024, p = 0.0028).In addition to these main effects, we uncovered several significant interaction effects.The interaction between group and condition proved to be statistically significant (F(2, 34) = 3.996, p = 0.0190), suggesting that the influence of condition on our outcome might vary across different groups.Similarly, we found a significant interaction between group and anaphora (F(1, 34) = 28.393,p = 1.49e-07), indicating that the effect of anaphora types on our response might differ depending on the group.Furthermore, the interaction between condition and anaphora emerged as highly significant (F(2, 34) = 32.703,p = 4.29e-14).Lastly, a significant three-way interaction among group, condition, and anaphora was observed (F(2, 34) = 15.608,p = 2.63e-07).We then computed multiple pairwise t-tests to calculate pairwise comparisons between groups with corrections for multiple testing.Among other significant differences, the results confirmed significant differences between children and adults in the partly match condition for anaphora modified by size adjectives.
Regarding a possible difference between anaphoric reconstruction in DP-coordination ellipsis and in sentence coordination ellipsis, for each condition, we fitted a model with Answer as a DV and Anaphora (Color vs Size one), Group (Children vs. Adults), and Sentence (Same/Different) as fixed effects.The model revealed no effect of Sentence, no interaction between Group and Sentence, no interaction between Anaphora and Sentence, and no interaction between Group, Anaphora and Sentence.Thus, there seems to be no difference between DP-coordination ellipsis and sentence coordination ellipsis.However, a study with more statistical power and more items per condition might reveal a different picture.4. Discussion.Our findings show that children have the ability to reconstruct unul/una as anaphoric rather than deictic as early as age 4.This is in line with our expectations given that Tasinato & Sanfelici's (2022) study showed that deictic errors tended to occur primarily with 3year-olds.
Interestingly, while children are able to reconstruct unul/una as an anaphor, they seem to resort to different anaphoric reconstruction strategies than adults.Adults' anaphoric reconstruction is consistently sensitive to nesting.Adults seem to prefer maximal reconstruction, reconstructing to the highest element under the AP that is substituted: to the NP măr verde 'green apple' in Ursuleţul vrea un măr verde mic şi unul mare 'The bear wants a small green apple and a big one' and to the N măr 'apple' in Ursuleţul vrea un măr verde mic şi unul galben 'The bear wants a small green apple and a yellow one'.Contrastively, except for two children who are adult-like, most of the children (N=16) differ from adults in their anaphoric reconstruction.Four children reject unul 'one' as a referent to a noun modified by a different property (yellow apple, big apple), reconstructing to an NP/N' that includes the antecedent and the previously mentioned property: the color property in the case of anaphors modified by Size adjectives, the size property in the case of anaphors modified by Color adjectives.Seven children accept unul 'one' to refer to different properties in both structures, operating in terms of minimal reconstruction, therefore avoiding adjective recursion.Reconstruction to N leaves the other properties unspecified, allowing children to express their own preference (possibly influenced by pragmatic biases).The remaining children reconstruct both minimally and maximally.The results thus show that children's reconstruction preferences are different from adults.This may suggest that children do not nest adjectives in the same hierarchical manner as adults.Instead, they could treat the Color and Size adjectives as coordinated such that they are both reconstructed together alongside N (but one of the properties is substituted), or both ignored in favor of the simpler N. Another possibility is that children may nest adjectives hierarchically but they do not have a fixed preferential hierarchy at this stage, i.e. they place the Color adjective closer to the noun than the Size adjective or they place the Size adjective closer to the noun than the Color adjective.If this is the case, then children who prefer to keep the same property as mentioned before could be argued to operate in terms of maximal reconstruction to the NP/N' nested immediately under the adjective modifying the anaphor.
Moreover, it is not excluded that other biases may interfere in anaphoric reconstruction.For instance, it could be that while doing reconstruction to N some children nonetheless prefer to keep previously mentioned properties in virtue of a property conservation cognitive bias.
While our experiment cannot untangle the variety of possible theoretical accounts, overall, the data from Romanian 4-year-olds suggests a hybrid stage in anaphoric reconstructions, where multiple reconstruction strategies are used.This is in contrast to the adult data, showing a consistent preference for maximal reconstruction that is constrained by hierarchical structure.

Figure 4 .
Figure 4. Examples of pictures tested for (9) 3.3.PROCEDURE.We employed a truth value judgment task containing 22 picture sets and utterances: 2 warm-ups, 4 controls and 16 test sentences.Participants heard an animal's wish (i.e., what the animal wanted) and evaluated whether a boy called Mihai gave it the right object.The warm-up sentences presented participants with two situations: (i) a situation where the boy gave the character the right object (see (14a), where the boy gives the Gorilla a banana), and (ii) a situation where the boy gave the character the wrong object (see (14b), where the boy gives the Kitty a bunny instead of a ball).This procedure familiarizes participants with both types of situations they encounter throughout the experiment.(14) Warm up sentences: a. Maimuţa vrea o banană şi Gorila vrea o banană.Monkey wants a banana and Gorilla wants a banana 'Monkey wants a banana and Gorilla wants a banana.' b.Cuţu vrea un os şi Pisicuţa vrea o minge.Doggie wants a bone and Kitty wants a ball 'Doggie wants a bone and Kitty wants a ball.'
Look! Mihai gave the bear a small green apple.What do you think Mihai should give the bear now?Mihai wants to give him this! (see arrow) Is this what the bear wants?What would you call this?If answer is yes, continue to the next slide.If answer is no, ask: What do you think the bear wants?

Figure 6 .
Figure 6.Acceptance rates per Condition and Anaphora across Groups

Table 1 .
Hierarchical representations of nouns and anaphors modified by a Color and/or a Size adjective